When it comes to vibe coding... Fast is slow. And slow is fast.

Why is Bluesky so anti AI? I don’t get it. A brand offers a privacy-friendly way to use AI… And most comments just hate it because it’s AI.

Has anyone had success with using swarms of agents, taskmaster etc with Claude Code? Every time I try, I find that giving small, well-defined tasks (aka going slow) beats going fast every time.

Is Test-Driven Development still the best approach when using Claude Code or is it now outdated?

Be honest, does this video suck? I made it quickly before Lorelight's PH launch. To me it sounds kind of embarrassing. My gf thinks it shows a real human touch. The one and only @viperchill.bsky.social thinks my accent sounds cool. So what do you all think? www.loom.com/share/aab1ad...

Lorelight, my SaaS that helps you monitor your brand's AI narrative and find hidden mentions is live on Product Hunt and I would love your support! Could you please click on the link in the comments and check it out! It would mean a lot to me :)

Claude Code fits my workflow perfectly because I spend 90% of my time reading research papers about machine learning and architecting and only 10% implementing. Claude allows me to validate and disprove hypotheses 10 times faster.

Bluesky friends, I need your help! Lorelight is now live on tinystartups.com and it needs to reach the #1 spot! Please upvote and comment on this post for reach! Merci !

I don't know why people say building is the easy part. Even with amazing tools like Claude Code, building Lorelight.ai and getting it to reliably handle entity resolution, enrichment, local AI models etc, has been way harder than marketing so far.

Claude Code is the best thing that ever happened to software engineering. Change my mind.

Between the new DeepSeek R1, Claude 4 etc, there has been a lot of new AI models lately. What is everyone using for coding? I currently use Claude 4 Opus for initial plans, then Sonnet for implementation. But even that is getting expensive. Anyone tried the new DeepSeek?

Me: "I'm tired of language learning, too much competition." Also me: Create Lorelight.ai, a B2B app in one of the hottest niches of 2025: AI search optimization.

I love Zed and @anthropic.com but Claude 4 Sonnet is getting really expensive. Any alternative, almost as good models, that happen to be cheaper? Not necessarily for all tasks but maybe for easier tasks.

Bluesky friends, I need your help! Lorelight, my SaaS helping you monitor your visibility and reputation in ChatGPT is launching on ProductHunt next week. Can you help me spread the work by clicking on the link in bio and clicking "Notify Me"? Merci! :)

Claude 4 just changed my admin password and just prevented me from accessing my own app's admin interface. It's true what they say. AI truly is taking over.

Shipped Lorelight.ai yesterday, and I already have a company eager to try it. During onboarding, I immediately realized I’d neglected one feature and spent too much time on others. That’s why shipping fast is essential, you don’t know what your users will actually care about until you ship.

After 12 years running a B2C SaaS (and plenty of complaining 😅), I finally shipped my first B2B product: lorelight.ai It tracks what ChatGPT, Claude & Gemini say about your brand—so you can catch risks early. Early access is limited to 20 companies. Don’t miss out.

One day before launch, I ripped out half my app’s code. Why? It was too complex, too soon. I want to ship a genuine MVP, learn from real users, then iterate, instead of rolling out a bloated solution I can’t scale.

The AI coding agent rollercoaster. "You're absolutely right. Let's simplify and follow exactly what the documentation states, without any custom files or copying:" 1 minute later: "OK I disregarded everything the docs says and built a custom implementation just for fun."

Vibe coding during the weekend is great. Way fewer overloaded errors from Claude and ChatGPT. Highly recommend.

I don't get people who say AI is not that great. I have built my dream product in 3 weeks start to finish and learned a ton in the process. This would have taken me months without AI. Honestly, it still blows my mind.

Haven't even launched and I have already been contacted by multiple companies eager to try @LorelightAI . Feeling grateful and excited to share it with everyone soon!

Indie hackers are the best. I shared my plan to launch my new SaaS in one week and got so many words of encouragement and people cheering me on. Thank you!

Bluesky, I need your help! My new SaaS Lorelight is almost there (see my profile for more info). Except it has been almost there for a while now... And I want to finally ship it. Can you hold me accountable and make sure I ship on Wednesday next week at the latest?

I feel like many of you have been quiet lately. What are you all building? Personally, I have decided to move away from language learning and build lorelight.ai. Launching soon.

There is a middle ground between shipping a garbage product and never shipping. Shipping fast doesn't mean releasing a bad product, it means releasing an imperfect product you can refine based on customer feedback. Oh and it allows you to not go bankrupt in the process.

Back to work after being away for a while. Which AI models are you all using for coding?

Are people actually getting good results using Gemini 2.5 Pro for coding? Every time I use it, it hallucinates so badly I can't get it to do anything even remotely useful. Makes me wonder if all the amazing feedback is real.

Loving @zed.dev ‘s new agentic editing, Built half of my MVP in just a few hours with it. Hoping to launch by the end of the week.

B2C customer feedback of the week when asking why they cancelled: “It costs something.”

Do I know anyone in PR/comms? Would love to have a quick call about your use of AI. No pitching, just looking to validate an idea.

Ok I give up. I thought I could use Lovable or Replit Agent to quickly create a MVP and validate ideas… But I keep going around in circles. Will code the app myself instead.

Them: just let your AI agent code while you sleep. The AI agent: I decided to switch to a completely different auth library just for fun. Oh and look at all the cool features I implemented even though nobody requested them.

One thing I have noticed when vibe coding: Asking another AI to rewrite prompts always leads to way better results.

All these people saying “vibe coding” doesn’t work… Just gave it a try on Replit and had a working SaaS MVP in a week. It’s not perfect, but it’s live, real people are using it, and I’m validating much faster than if I had built it myself.

I’ve built an MVP for a B2B SaaS (no auth or payments yet). Do I: 1) Add auth + Stripe, then cold email/run ads? 2) Talk to prospects without pitching to validate the problem first?

Vibe coding is such an emotional rollercoaster.

My vibe coding experience. Replit Agent v2 easily built a MVP but never managed to set up authentication. Lovable easily set up authentication but took way more tries to build a functional MVP.

What's the best way to validate in the age of vibe coding? 1) Calls with potential prospects then code MVP 2) Basic research then use @Replit or @lovable_dev to make a MVP Tempted to go with #2 and then code the app by myself if there is interest.

Mixed feelings about Deep Research after using it dozens of times. The main issue is that it often hallucinates sources, stats and quotes. Meaning you end up with conclusions that are based on fake data and therefore most likely incorrect.

Building a SaaS to solve a problem I suspect agencies face. But before building it, I want to make sure it's an actual problem. If you work at a branding/design agency, would you mind taking a few minutes to help me find out? No pitch, just validation.

Agency owners: How many times have you lost a $10K+ naming/branding project because stakeholders "couldn't visualize" the concept? Would love a 15-min call to learn about your experience. Building something to solve this in 60 seconds with AI. No pitch - just learning. DM if interested 🙏

Launching and validating an idea for a SaaS helping marketing agencies close deals. .com costs 3k, .ai and .io are available. GetSaaSName.com also available. Which domain do you go for?

Claude 3.7 sonnet is like that super smart friend who occasionally says genius things but spends 90% of the time going on wild tangents. You never know what’s next.

Mixed feelings about Claude 3.7 Sonnet. I spent all day using it for a mix of marketing research and coding. It does seem smarter but it also makes more mistakes and frequently forgets what its doing. Will keep using 3.5 for prompts requiring lots of output.

Looking into possibly buying a B2B SaaS. Anyone selling? Max budget is 20 - 30k.

Trying to decide whether I should pivot my existing B2C SaaS, create another B2C SaaS with different positioning… Or just validate B2B SaaS ideas and say bye to B2C. Tough choice.

I have a B2B SaaS idea I’m super excited about. - solves a real, quantifiable problem - solution feels like magic and value is obvious - can build a MVP in 2 weeks Will start validating next week and set up a few calls.

Considering two different positioning angles for my new SaaS Copycat Cafe. 1) Build Your Personal French Phrase Library - frenchtogether.com/library/ 2) Steal Native French Speakers' Exact Words - frenchtogether.com/copy/ Which resonates most with you and why?

I hate boring brands. I want a brand brimming your personality. The kind of brand some people will hate and others will love. Working on that for my next SaaS :)